Ling 566 Nov 28, 2011 Long Distance Dependencies
Overview Some examples of the phenomenon What is new and different about it Brief sketch of the TG approach Broad outlines of our approach Details of our approach Subject extraction Coordinate Structure Constraint
Examples wh-questions: What did you find? Tell me who you talked to relative clauses: the item that I found the guy who(m) I talked to topicalization: The manual, I can t find Chris, you should talk to. easy-adjectives: My house is easy to find. Pat is hard to talk to.
What these have in common There is a gap : nothing following find and to, even though both normally require objects. Something that fills the role of the element missing from the gap occurs at the beginning of the clause. We use topicalization and easy-adjectives to illustrate: The manual, I can t find Chris is easy to talk to
Gaps and their fillers can be far apart: The solution to this problem, Pat said that someone claimed you thought I would never find. Chris is easy to consider it impossible for anyone but a genius to try to talk to. That s why we call them long distance dependencies
Fillers often have syntactic properties associated with their gaps Him, I haven t met. *He, I haven t met. The scissors, Pat told us were missing. *The scissors, Pat told us was missing. On Pat, you can rely. *To Pat, you can rely.
LDDs in TG These were long thought to constitute the strongest evidence for transformations. They were handled in TG by moving the filler from the gap position. Case, agreement, preposition selection could apply before movement.
A big debate about LDDs in TG Does long-distance movement take place in one fell swoop or in lots of little steps? Swooping Looping
Looping is now generally accepted in TG Various languages show morphological marking on the verbs or complementizers of clauses between the filler and the gap. Psycholinguistic evidence indicates increased processing load in the region between filler and gap. This opens the door to non-transformational analyses, in which the filler-gap dependency is mediated by local information passing.
Very Rough Sketch of Our Approach A feature GAP records information about a missing constituent. The GAP value is passed up the tree by a new principle. A new grammar rule expands S as a filler followed by another S whose GAP value matches the filler. Caveat: Making the details of this general idea work involves several complications.
The Feature GAP Like valence features and ARG-ST, GAP s value is a list of feature structures (often empty). Subject gaps are introduced by a lexical rule. Non-subject gaps are introduced by revising the Argument Realization Principle.
word: The Revised ARP SYN VAL GAP ARG-ST A B [ SPR A COMPS B C C ] is a kind of list subtraction, but: it s not always defined, and when defined, it s not always unique The ARP now says the non-spr arguments are distributed between COMPS and GAP.
A Word with a Non-Empty GAP Value hand, word SYN ARG-ST [ HEAD FORM VAL GAP 2 NP[acc] ] fin [ SPR 1 COMPS 3 PP[to] [ 1 NP ] CASE nom, 2, 3 AGR non-3sing ]
How We Want GAP to Propagate [ S ] GAP [ NP ] [ S ] GAP GAP NP Kim [ NP ] [ VP ] GAP GAP NP we [ V ] [ S ] GAP GAP NP know [ NP ] [ V(P) ] GAP GAP NP Dana hates
What We Want the GAP Propagation Mechanism to Do Pass any GAP values from daughters up to their mothers, except when the filler is found. For topicalization, we can write the exception into the grammar rule, but For easy-adjectives, the NP that corresponds to the gap is the subject, which is introduced by the Head-Specifier Rule. Since specifiers are not generally gap fillers, we can t write the gap-filling into the HSR.
Our Solution to this Problem For easy-adjectives, we treat the adjective formally as the filler, marking its SPR value as coindexed with its GAP value. We use a feature STOP-GAP to trigger the emptying of the GAP list. STOP-GAP stops gap propagation easy-adjectives mark STOP-GAP lexically a new grammar rule, the Head-Filler Rule mentions STOP-GAP
The GAP Principle A local subtree Φ satisfies the GAP Principle with respect to a headed rule ρ if and only if Φ satisfies: [ GAP ( A 1... A n ) A 0 ] [GAP A 1 ]... [ GAP H A i STOP-GAP A 0 ]... [GAP A n ]
How does STOP-GAP work? STOP-GAP is empty almost everywhere When a gap is filled, STOP-GAP is nonempty, and its value is the same as the gap being filled. This blocks propagation of that GAP value, so gaps are only filled once. The nonempty STOP-GAP values come from two sources: a stipulation in the Head-Filler Rule lexical entries for easy-adjectives No principle propagates STOP-GAP
The Head-Filler Rule [ ] [phrase] 1 GAP HEAD H VAL [ verb FORM [ fin ] SPR COMPS STOP-GAP 1 GAP 1 ] This only covers gap filling in finite Ss The filler has to be identical to the GAP value The STOP-GAP value is also identical The GAP Principle ensures that the mother s GAP value is the empty list
Gap Filling with easy-adjectives easy, adj-lxm SYN ARG-ST [ ] STOP-GAP 1 [ VP ] NP i, INF + GAP 1 NP i,... Because STOP-GAP and GAP have the same value, that value will be subtracted from the mother s GAP value. The first argument is coindexed with the GAP value, accounting for the interpretation of the subject as the filler.
A Tree for easy to talk to [ ] VAL SPR 2 NP i GAP [ A ] SPR 2 VAL COMPS 3 GAP STOP-GAP 1 easy 3 [ VP ] VAL SPR NP GAP 1 NP i to talk to
STOP-GAP Housekeeping Lexical entries with nonempty STOP-GAP values (like easy) are rare, so STOP-GAP is by default empty in the lexicon. Head-Specifier and Head-Modifier rules need to say [STOP-GAP < >] Lexical rules preserve STOP-GAP values.
GAP Housekeeping The initial symbol must say [GAP < >]. Why? To block *Pat found and *Chris talked to as stand-alone sentences. The Imperative Rule must propagate GAP values. Why? It s not a headed rule, so the effect of the GAP Principle must be replicated Imperatives can have gaps: This book, put on the top shelf!
Sentences with Multiple Gaps Famous examples: This violin, sonatas are easy to play on. *Sonatas, this violin is easy to play on. Our analysis gets this: The subject of easy is coindexed with the first element of the GAP list. The Head-Filler rule only allows one GAP remaining. There are languages that allow multiple gaps more generally.
Where We Are filler-gap structures: The solution to this problem, nobody understood That problem is easy to understand The feature GAP encodes information about missing constituents Modified ARP allows arguments that should be on the COMPS list to show up in the GAP list GAP values are passed up the tree by the GAP Principle
Where We Are (continued) The feature STOP-GAP signals where GAP passing should stop The Head-Filler Rule matches a filler to a GAP and (via STOP-GAP) empties GAP Lexical entries for easy-adjectives require a gap in the complement, coindex the subject with the gap, and (via STOP-GAP) empty GAP on the mother
On to New Material. Sentences with subject gaps Gaps in coordinate constructions
Subject Gaps The ARP revision only allowed missing complements. But gaps occur in subject position, too: This problem, everyone thought was too easy. We handle these via a lexical rule that, in effect, moves the contents of the SPR list into the GAP list
The Subject Extraction Lexical Rule NB: This says nothing about the phonology, because the default for pi-rules is to leave the phonology unchanged. pi-rule INPUT X, SYN HEAD [ verb FORM fin ] VAL [ SPR Z ] ARG-ST A OUTPUT Y, SYN VAL [ SPR ] GAP 1 ARG-ST A 1,...
A Lexical Sequence This Licenses word [ ] verb HEAD FORM fin [ ] SPR likes, VAL SYN COMPS 2 [ ] CASE nom GAP 1 AGR 3sing STOP-GAP ARG-ST 1, 2 NP[acc] Note that the ARP is satisfied
A Tree with a Subject Gap [ S ] GAP [ NP ] [ S ] GAP GAP NP Kim [ NP ] [ VP ] GAP GAP NP we [ V ] [ S ] GAP GAP NP know [ V ] NP GAP NP likes Dana
Island Constraints There are configurations that block filler-gap dependencies, sometimes called islands Trying to explain them has been a central topic of syntactic research since the mid 1960s We ll look at just one, Ross s so-called Coordinate Structure Constraint Loose statement of the constraint: a constituent outside a coordinate structure cannot be the filler for a gap inside the coordinate structure.
Coordinate Structure Constraint Examples *This problem, nobody finished the extra credit and *This problem, nobody finished and the extra credit. *This problem, nobody finished and started the extra credit. *This problem, nobody started the extra credit and finished But notice: This problem, everybody started and nobody finished
The Coordinate Structure Constraint In a coordinate structure, no conjunct can be a gap (conjunct constraint), and no gap can be contained in a conjunct if its filler is outside of that conjunct (element constraint)..unless each conjunct has a gap that is paired with the same filler (across-the-board exception)
These observations cry out for explanation In our analysis, the conjunct constraint is an immediate consequence: individual conjuncts are not on the ARG-ST list of any word, so they can t be put on the GAP list The element constraint and ATB exception suggest that GAP is one of those features (along with VAL and FORM) that must agree across conjuncts. Note: There is no ATB exception to the conjunct constraint. *This problem, you can compare only and.
Our Coordination Rule, so far FORM 1 VAL 0 IND s 0 FORM 1 FORM 1 VAL 0... VAL 0 IND s 1 IND s n 1 HEAD conj IND s 0 [ ] RESTR ARGS s 1...s n FORM 1 VAL 0 IND s n Recall that we have tinkered with what must agree across conjuncts at various times. Now we ll add GAP to the things that conjuncts must share
Our Final Coordination Rule FORM 1 VAL 0 GAP A IND s 0 FORM 1 VAL 0 GAP A... IND s 1 FORM 1 VAL 0 GAP IND A s n 1 HEAD conj IND s 0 [ ] RESTR ARGS s 1...s n FORM 1 VAL 0 GAP IND A s n We ve just added GAP to all the conjuncts and the mother. This makes the conjuncts all have the same gap (if any) Why do we need it on the mother?
Closing Remarks on LDDs This is a huge topic; we ve only scratched the surface There are many more kinds of LDDs, which would require additional grammar rules There are also more island constraints, which also need to be explained Our account of the coordinate structure constraint (based on ideas of Gazdar) is a step in the right direction, but it would be nice to explain why certain features must agree across conjuncts.
Overview Some examples of the phenomenon What is new and different about it Brief sketch of the TG approach Broad outlines of our approach Details of our approach Subject extraction Coordinate Structure Constraint
Reading Questions Where do non-empty STOP-GAP values come from? Where do empty STOP-GAP values come from? Is STOP-GAP only non-empty on the head daughter of HFR? What s the purpose of STOP-GAP?
Reading Questions When the HFR says the head has nonempty GAP and STOP-GAP, does that mean it both has a gap and fills a gap? In the easy to talk to tree, what rule licenses the phrase?
[ ] [phrase] 1 GAP HEAD H VAL [ verb FORM [ fin ] SPR COMPS STOP-GAP 1 GAP 1 ] [ ] VAL SPR 2 NP i GAP [ A ] SPR 2 VAL COMPS 3 GAP STOP-GAP 1 easy 3 [ VP ] VAL SPR NP GAP 1 NP i to talk to
Reading Questions In (38) on p. 439, why is the SPR of the VP to talk to an NP rather than the A easy? Also, if the element [1]NPi in the GAP list of the VP to talk to and the SPR of the AP easy to talk to ([2]NPi) are coindexed, why aren't they co-tagged with the same number as well?
Reading Questions How is using the GAP feature any different than using an "empty element" like other grammars use?